[PATCH] D24623: AMDGPU: Implement memory model

[Tony Tye via Phabricator via llvm-commits]

t-tye added inline comments.


================
Comment at: lib/Target/AMDGPU/SIMemoryLegalizer.cpp:327
+  AtomicOrdering SuccessOrdering = MMO->getOrdering();
+  AtomicOrdering FailureOrdering = MMO->getFailureOrdering();
+  SyncScope::ID SSID = static_cast<SyncScope::ID>(MMO->getSyncScopeID());
----------------
May be worth asserting that FailureOrdering is not AtomicOrdering::Release or AtomicOrdering::AcquireRelease as these are not allowed, and following code relies on that fact.


================
Comment at: test/CodeGen/AMDGPU/flat_atomics.ll:1001
 ; GCN-LABEL: {{^}}atomic_store_i32_offset:
-; CIVI: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}} glc{{$}}
-; GFX9: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}} offset:16 glc{{$}}
+; CIVI: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}}{{$}}
+; GFX9: flat_store_dword v[{{[0-9]+}}:{{[0-9]+}}], {{v[0-9]+}} offset:16{{$}}
----------------
Curious why glc is no longer being checked for?


https://reviews.llvm.org/D24623